yolov5进行分类模型训练

数据集收集

datasets

train

classes1_imgs

classes2_imgs

classes3_imgs

…

test

classes1_imgs

classes2_imgs

classes3_imgs

…

valid

classes1_imgs

classes2_imgs

classes3_imgs

…

建议数据集划分为 train: test: valid = 7: 2: 1

训练模型

下载并配置yolov5环境

需要在pytorch gpu版本下进行

git clone https://github.com/ultralytics/yolov5 
cd yolov5
pip install -r requirements.txt  # 下载所需包

训练模型

将数据集放在 yolo5 目录下即可

打开train.py, 设置以下四个参数

parser.add_argument('--model', type=str, default='yolov5s-cls.pt', help='initial weights path')
    parser.add_argument('--data', type=str, default='imagenette160', help='cifar10, cifar100, mnist, imagenet, ...')
    parser.add_argument('--epochs', type=int, default=10, help='total training epochs')
    parser.add_argument('--batch-size', type=int, default=64, help='total batch size for all GPUs')
    parser.add_argument('--imgsz', '--img', '--img-size', type=int, default=224, help='train, val image size (pixels)')

–data 为数据集路径
–batch-size 为每个训练批次的图片数, 提高该数值会增加训练速度,同时也会带来更高的内存占用
–epochs 训练次数

模型默认使用yolov5s-cls.pt, 各个预训练模型参数如下

之后运行 train.py 训练模型

使用模型

from utils.augmentations import classify_transforms
from utils.dataloaders import LoadImages
from utils.general import Profile, check_img_size, non_max_suppression, scale_boxes
from models.common import DetectMultiBackend
from utils.torch_utils import select_device

import pathlib
temp = pathlib.PosixPath
pathlib.PosixPath = pathlib.WindowsPath

import torch
import torch.nn.functional as F


def predict_oneImg(model, dir_path, imgsz = (224, 224), dt = (Profile(), Profile(), Profile())):
    dataset = LoadImages(dir_path, img_size=224,
                         transforms=classify_transforms(imgsz[0]))
    for path, im, im0s, vid_cap, s in dataset:
        with dt[0]:
            im = torch.Tensor(im).to(model.device)
            if len(im.shape) == 3:
                im = im[None]  # expand for batch dim

        # Inference
        with dt[1]:
            results = model(im)

        # Post-process
        with dt[2]:
            pred = F.softmax(results, dim=1)

        classes_names = model.names
        prob_list = pred.tolist()[0]
        top3pre = pred.argsort(descending=True).tolist()[0][:3]

        for i in top3pre:
            print("{}: {:.3}%".format(classes_names[i], prob_list[i] * 100), end=" ")
        print('')


device = ''
device = select_device(device)
# 模型路径
model_cls_o = DetectMultiBackend('weights/ball_card_cls2optic.pt', device=device)
# 图片路径
predict_oneImg(model_cls_o, dir_path=r"D:\Code\ML\images\Mywork2\Url_O\16-17")